[spec] Rework and correct explanation of deduplication tables.

This commit is contained in:
2025-08-01 04:18:10 +01:00
parent ff1e949346
commit ac92c6d3d3

View File

@@ -121,60 +121,77 @@ The size type defines the following type of entries:
|Each entry uses five bytes, with the leftmost byte used for flags and the next three bytes used as a pointer to the sector or next level. |Each entry uses five bytes, with the leftmost byte used for flags and the next three bytes used as a pointer to the sector or next level.
|=== |===
==== Sector Pointer Resolution and Table Levels ==== Interpretation of Deduplication Table Entries
When `levels` is equal to 1—indicating a single-level deduplication table—each entry in the table corresponds directly to a media sector. Decoding deduplication tables may seem complex initially, but the logic is structured and manageable.
The pointer value is resolved using the following procedure: Three parameters are critical for interpreting deduplication table entries:
- Right-shift the raw pointer value by the `shift` value. - *block_alignment_shift*
- Multiply the result by the `alignment` to compute the absolute byte offset of the target data block. - *table_shift*
- The remainder of the original pointer value modulo `(1 << shift)` yields the item index within the block. - *data_shift*
Each data block stores a fixed number of bytes per sector, allowing compact and efficient sector addressing. These parameters are stored in both the master header and each deduplication table header to support reliable decoding.
_For example_: ===== Block Alignment
Given a pointer value of `0x8003`, a `shift` of 5, and an `alignment` of 9:
- `0x8003 >> 5 = 0x400 = 1024`
- `1024 * 9 = 9216`
- The sector index within the block is `0x8003 & 0x1F = 3`
Thus, the sector is located at byte offset `9216`, and it is the 3rd item in the block. Each block in the image is aligned to a boundary of `2 << block_alignment_shift`.
This alignment is essential for technical consistency and performance.
===== Multi-Level Tables ===== Table Shift
When `levels > 1`, the interpretation of pointer entries changes substantially. The `table_shift` parameter defines how many blocks (or sectors) are represented by each entry, based on the deduplication table level.
Although typical usage involves no more than two levels, implementations **MUST** be capable of handling an arbitrary number of levels to ensure forward compatibility. In multi-level tables, this value governs an exponential reduction in scope per level.
At each level—except the final—the table entry functions as an address to the next-level table. For example:
The range of LBAs covered by each entry is calculated as:
[cols="1,2",options="header"]
|===
| Level
| Sectors per Entry
| 1
| (2 << table_shift)^2 = 262144
| 2
| 2 << table_shift = 512
| 3
| 1
|===
Tables with more than two levels are rare, but implementations should be resilient enough to handle unexpected depths gracefully.
===== Entry Format Across Levels
In non-terminal levels (i.e., all except the last), each entry contains:
- Relevant metadata flags for its sector range
- An offset pointing to the next deduplication level
To obtain the byte offset in the image file, multiply this offset by `2 << block_alignment_shift`.
In the last level, the `data_shift` is applied as follows to determine the specific item within a data block:
.Example calculation
[source] [source]
range = entry_index * (1 << shift)^(levels - 1) ----
Given:
- Entry value = 0x35006
- data_shift = 5
- block_alignment_shift = 9
_For example_, with a `shift` value of 9 and two levels: Step 1: Mask and shift
- Entry `0` spans LBAs `0511` 0x35006 >> 5 = 0x1A80
- Entry `1` spans LBAs `5121023`
With three levels: Step 2: Compute byte offset
- Entry `0` at level 0 spans LBAs `0262143` 0x1A80 * (2 << 9) = 0x6A0000
- Entry `0` at level 1 within that region spans LBAs `0511`, and so on recursively.
===== Resolution Example Step 3: Determine item index
0x35006 & 0x1F = 6
To locate sector `1012` using a two-level table with `shift = 9` and `alignment = 9`: Result:
Sector is stored at byte offset 0x6A0000 as item number 6 in the data block.
1. **Level 0**: ----
- Sector `1012` falls within entry `1` (covers `5121023`)
- Entry `1` contains the value `0x12000`
- Multiply by `alignment` → `0x12000 * 9 = 0x225000 = 37,748,736`
- Read the next-level table at byte offset `37,748,736`, marked with the identifier `DDTS`
2. **Level 1**:
- The relevant entry is `500` (`1012 - 512 = 500`)
- Entry `500` contains `0x35006`
- Right-shift `0x35006 >> 9 = 0x6A = 106`
- Multiply by `alignment`: `106 * 9 = 954`
- Sector resides at byte offset `217,088` and is the 6th item in the block (`0x35006 & 0x1FF = 6`)
===== Deduplication table flags ===== Deduplication table flags